California panel report targets RSI risks: compute disclosure, SL3 security, outcome standards
gleech · x · 2026-11-20
A California governor's panel report drawing attention from AI policy figures proposes requiring companies to disclose what share of compute goes to frontier AI R&D, outcome-based standards complementing safety cases, and upgrading labs to SL3+ model weight security. Fabien Roger argues limiting compute spent on AI R&D is among the most viable paths to pacing the frontier; Sneha Revanur calls it the world's top near-term policy opportunity for managing RSI risk, with parts potentially becoming law within months.
Related event: California Governor's AI Report Puts Recursive Self-Improvement First(2 posts)→
More from Safety
- Hacker Swapped API's DeepSeek for Abliterated Version, Undetected for Two Days — moyix · 2026-11-20
- Claim: METR is an Anthropic-linked EA arm, not an independent evaluator — gerardsans · 2026-11-20
- Fuck Your AI Safeguards: a hands-on look at bypassing AI guardrails — speckx · 2026-11-20
- US Senator's open letter calls for banning recursive self-improvement in AI — S_OhEigeartaigh · 2026-11-20
- Luiza Jarovsky: AI's major transparency problem amid rising AI cyberattacks — LuizaJarovsky · 2026-11-20
- Cross-institutional survey asks whether mechanistic interpretability actually makes AI safer — sarahookr · 2026-11-20